Jeremy Neely Text Jeremy

AI & Agents · Production

AME MCP Server

One governed doorway between AI assistants and business systems.

Built for: The AME Group, the managed IT provider I work for, plus client-facing deployments for a truck dealer and an AI tips service

How the gateway sits between AI clients and business systems.
How the gateway sits between AI clients and business systems.

The problem

AI assistants are only useful if they can read real data such as customers, units and tickets. The easy way is to paste a business system's API key into the AI platform, which leaves a credential in a third-party system, no record of what the AI asked for, and every integration breaking separately when a vendor changes its API.

What I built

I built a Model Context Protocol server that sits in the middle. Each AI consumer gets a narrow key that only sees its own tools, the server holds the real credentials, and every call is logged. Tools are imported from OpenAPI, Postman, GraphQL, SOAP or cURL definitions, and the server is documented against a formal security architecture.

How it works

  1. Import a business API from its spec, or build a tool with the AI Tool Builder.
  2. Give each AI platform its own scoped key.
  3. The AI calls a tool; the server checks the key, validates the arguments and adds the real credential.
  4. The server calls the business system and returns only what the tool allows.
  5. Every call is written to a structured audit log.

Built with

  • Python
  • FastAPI
  • Model Context Protocol
  • OAuth 2.1 / PKCE
  • pydantic
  • pytest

More of Jeremy's work